What Is the IQ of ChatGPT? Scores, Tests and Limits
What is the IQ of ChatGPT? There's no official number. Early 2023 tests of ChatGPT on the verbal part of human IQ tests reported scores around 147 to 155, and newer models score higher on puzzle-style tests every few months. But those numbers come from tests built for humans, so a "ChatGPT IQ" says more about the test than about how the AI thinks.
If you're asking because you've started outsourcing every hard question to a chatbot and you're lowkey wondering what your own brain scores, IQ Tester shows which kinds of reasoning you're strong in and where you slip: Find my weak spots.
What is ChatGPT's IQ, really?
Short version: it depends on which test, which model, and who ran it.
Here are the results that get passed around the most, stated as carefully as we can:
- Verbal IQ, early 2023. Psychologist Eka Roivainen wrote in Scientific American that he gave ChatGPT the verbal sections of a standard adult IQ test and it scored about 155 on verbal IQ.
- Another verbal test, early 2023. Researcher David Rozado wrote on his Substack that ChatGPT scored around 147 on a verbal-linguistic IQ test.
- Puzzle tests since then. Hobbyist trackers run newer models on public matrix-style quizzes (the kind with shape grids). Scores have climbed with nearly every release, and the "smartest AI model" changes hands often.
So when someone in the group chat says "ChatGPT has a 155 IQ," they're half right. That was one model, on one part of one test, years ago in AI time.
What about the ChatGPT 5 IQ score?
People search "what is the iq of chatgpt 5" constantly. We're not going to make up a number. Any score you see for a specific new model is usually one person's run on one online test, and those tests were never normed on AI. Treat it as a fun headline, not a fact.
How people test AI on IQ questions
There are basically three ways people run a ChatGPT IQ test.
- Paste in a real test. Someone copies verbal or number questions from an IQ-style test into the chat and scores the answers by hand. This is what the early 2023 verbal results did.
- Describe the pictures. Matrix puzzles are visual, so older models had them typed out in words ("top row: circle, square, triangle..."). Newer models can read images directly.
- Use a ChatGPT IQ test prompt. On Reddit, people share prompts like "Answer these 20 questions, show your reasoning, then give one final answer each." That's fine for vibes, but it's not a controlled test.
Why the AI IQ test results are shaky
| Problem | Why it messes up the score |
|---|---|
| Training data | The model may have seen the exact questions online already |
| No time limit | Human tests often time you; the AI doesn't feel pressure |
| Text vs pictures | Typing out a visual puzzle changes the puzzle |
| Norms | IQ scores compare you to people your age. An AI has no age group |
| One run | Ask twice and you can get different answers |
That last row is huge. An IQ score is a comparison to a crowd of humans. A chatbot isn't in that crowd, so its "score" is more like a translated grade than a real IQ.
Why AI IQ scores keep jumping
Every new model release, a new chatgpt iq estimate goes viral. Here's why the number keeps moving:
- Bigger, better-trained models. More training and better methods mean more puzzles solved.
- "Thinking" modes. Newer models can work through steps before answering, which helps a lot on number series and logic.
- Leaked questions. Once a test is public, it's probably in some training data. That inflates scores.
- Different testers. Reddit threads, blogs and trackers all use different questions, so their numbers aren't comparable.
If you've seen "chatgpt iq reddit" threads arguing 120 vs 150, this is why. They're often not even measuring the same thing.
Where AI still fails puzzles
AI is very good at verbal analogies and textbook logic. It's less reliable on things that look easy to you. Common weak spots people report:
- Counting things in a picture. Ask how many triangles are in a drawing and models sometimes guess. (Try our how many triangles puzzle answers yourself.)
- Trick questions that look familiar. If a puzzle looks like a famous riddle but has one detail changed, a model may answer the famous version.
- Brand-new puzzle types. Truly novel visual patterns are still harder for AI than for many humans.
- Knowing when it's wrong. It can sound super confident while being completely off.
That second one is basically the AI version of falling for puzzles that look easy but are actually hard. Humans do it too.
Is AI smarter than humans?
Depends what you mean by smart.
Where AI wins: speed, memory, vocabulary, reading a 50-page PDF in seconds, writing code that compiles.
Where you win: knowing what actually matters, noticing when a question is a trap, understanding the people in the room, and learning a new thing from one example without a billion training pages.
Which is smarter, Claude or ChatGPT?
Honest answer: it changes by the month and by the task. One model might be better at coding this season and worse at a certain kind of puzzle. Public leaderboards swap places all the time. If you want to compare, give both the same fresh questions they can't have memorized and check the reasoning, not just the final answer.
What a ChatGPT IQ score means for your own thinking
Here's the part nobody posts on TikTok. The real risk isn't that AI is smarter than you. It's that you stop practicing the stuff it does for you.
If ChatGPT always does the mental math, you get slower at mental math. If it always summarizes the reading, you get worse at pulling the main idea out yourself. That's not a scare story; it's just how skills work. Use it or lose it.
A simple rule for school, uni or work:
- Try first. Give a problem two honest minutes before asking AI.
- Ask for a hint, not the answer. "What's the first step?" keeps you in the driver's seat.
- Check its work. Spotting AI's mistakes is a real skill now.
- Notice what you always outsource. That's probably your weak spot.
The problem with step 4 is that you can't really judge your own blind spots. You think you're bad at math when it's actually reading the question that trips you up.
IQ Tester does that judging part for you. It mixes number series, verbal analogy, logic and pattern questions, starts easy and gets harder, and then shows how you did in each kind of reasoning. Then the guides give you one short drill a day on the weak one. Find my weak spots
Find out your own score in the next few minutes
Forget what is the IQ of ChatGPT for a sec and find out how your brain handles the same kinds of questions. IQ Tester gives you 20 untimed questions in your browser, and your band plus three of your four category strengths are free. You sign in with Google or Apple, no form. It's for fun, not a clinical test, but it shows you exactly where to practice.
FAQ
What is ChatGPT's IQ score in 2026?
There's no official chatgpt iq score. Early tests reported verbal IQ around 147 to 155, and newer models score higher on puzzle tests, but every result depends on the test and the model version. Treat any single number as an estimate, not a fact.
Can AI pass an IQ test?
Yes, modern AI can answer most questions on many IQ-style tests correctly, especially verbal ones. But an IQ score compares you to people your age, and an AI isn't part of that group. So it can pass the questions without the score meaning what it means for a human.
Is ChatGPT smarter than humans?
It's faster and knows more facts than any one person. It still makes confident mistakes, struggles with some brand-new visual puzzles, and doesn't understand context the way you do. Smarter at some tasks, not smarter overall.
Is there a ChatGPT IQ test prompt I can try?
Sure: give it 10 fresh puzzles it can't have seen, ask it to show reasoning and give one final answer each, then score it yourself. Run the same puzzles twice and you'll often see different answers, which is exactly why these results are shaky.
How can I test my own IQ online for free?
Plenty of quizzes exist, but many are clickbait. IQ Tester gives you 20 untimed reasoning questions, and your band and three of your four category strengths are free. It's for entertainment, not a clinical assessment.